Fir Moss Spider

Molecular Ecology (2015) 24, 3467–3484 doi: 10.1111/mec.13248

Sky island diversiﬁcation meets the multispecies coalescent – divergence in the spruce-ﬁr moss spider (Microhexura montivaga, Araneae, Mygalomorphae) on the highest peaks of southern Appalachia

MARSHAL HEDIN,* DAVE CARLSON* and FRED COYLE† *Department of Biology, San Diego State University, San Diego, CA 92182, USA, †PO Box 1935, Cullowhee, NC 28723, USA

Abstract Microhexura montivaga is a miniature tarantula-like spider endemic to the highest peaks of the southern Appalachian mountains and is known only from six allopatric, highly disjunct montane populations. Because of severe declines in spruce-ﬁr forest in the late 20th century, M. montivaga was formally listed as a US federally endangered species in 1995. Using DNA sequence data from one mitochondrial and seven nuclear genes, patterns of multigenic genetic divergence were assessed for six montane populations. Independent mitochondrial and nuclear discovery analyses reveal obvious genetic fragmentation both within and among montane populations, with ﬁve to seven primary genetic lineages recovered. Multispecies coalescent validation analyses [guide tree and unguided Bayesian Phylogenetics and Phylogeography (BPP), Bayes factor delimitation (BFD)] using nuclear-only data congruently recover six or seven distinct lineages; BFD analyses using combined nuclear plus mitochondrial data favour seven or eight lineages. In stark contrast to this clear genetic fragmentation, a survey of sec- ondary sexual features for available males indicates morphological conservatism across montane populations. While it is certainly possible that morphologically cryptic speciation has occurred in this taxon, this system may alternatively represent a case where extreme population genetic structuring (but not speciation) leads to an oversplitting of lineage diversity by multispecies coalescent methods. Our results have clear conservation implications for this federally endangered taxon and illustrate a methodological issue expected to become more common as genomic-scale data sets are gathered for taxa found in naturally fragmented habitats.

Keywords: allopatry, Bayes factor delimitation, conservation genetics, montane speciation, population structure, unguided Bayesian Phylogenetics and Phylogeography Received 26 January 2015; revision received 13 May 2015; accepted 14 May 2015

mountains (Hedin 1997; Weisrock & Larson 2006; Keith Introduction & Hedin 2012). In each of these areas, mountains act as Evolutionary biologists have long been interested in naturally fragmented habitat islands, serve as refugia in evolution on mountains. In mainland North America, the face of climatic variation and/or generate strong conspicuous hotspots for montane diversiﬁcation ecological gradients. These combinations of genetic and include the California Sierra Nevada (Rovito 2010; Scho- geographic isolation with potential selective differences ville & Roderick 2010; Hedin et al. 2013), ‘sky islands’ promote the evolution of both population genetic struc- of the desert southwest (Maddison & McMahon 2000; ture (arrays of geographically distinct populations Derkarabetian et al. 2011) and the southern Appalachian which are genetically divergent to various degrees) and clear species-level divergences. These naturally allopat- Correspondence: Marshal Hedin, Fax: +1 619 594 5676; ric systems also present classic difﬁculties for speciation E-mail: [email protected]

© 2015 John Wiley & Sons Ltd 3468 M. HEDIN ET AL. biologists – when allopatry prevails and population has been suggested to potentially oversplit diversity in structure is ubiquitous, distinguishing population sub- dispersal-limited taxa (e.g. Niemiller et al. 2012; Barley division from speciation is challenging and often ‘fuzzy’ et al. 2013; McKay et al. 2013; Satler et al. 2013). Authors (Leavitt et al. 2007; Bond & Stockman 2008; Keith & of these methods have acknowledged the inherent diffi- Hedin 2012; Satler et al. 2013). For example, Hey (2009) culties associated with fragmented, allopatric systems pointed out that the null hypothesis for many species (e.g. Zhang et al. 2013). delimitation methods is a ‘no significant differentiation’ Microhexura montivaga, the spruce-fir moss spider, is a model. Because both population subdivision and specia- miniature mygalomorph spider (F. Dipluridae) endemic tion imply differentiation (i.e. rejection of the null), such to high-elevation red spruce-Fraser fir forests of the methods can mistakenly equate these potentially differ- southern Appalachian mountains (Coyle 1981, 1985). ent evolutionary dynamics (see also Hey & Pinho 2012). Recent survey work (Coyle 2009) indicates that Micro- Modern researchers have access to many types of hexura is distributed as six disjunct montane popula- data when investigating the interface between populations, occupying the Virginia Balsam Mountains tion divergence and speciation, including data derived (Virginia), Roan Mountain (Tennessee/North Carolina), from morphology, behaviour, ecology and genomes. Grandfather Mountain (NC), the Black Mountains (NC), Assessing nuclear genomic divergence is particularly the Plott Balsam Mountains (NC) and the Great Smoky attractive in naturally fragmented systems because such Mountains (TN/NC) (Fig. 1). Most of these sky island data represent a common currency for measuring both populations are found above 1800 metres and include population genetic structure and speciation, and explicit the highest elevations in eastern North America (e.g. models are available that potentially distinguish popula- Mt. Mitchell, Black Mountains, NC). Spatial isolation tion structure from species-level divergence. These also potentially exists within mountain ranges, where include single-locus models (e.g. generalized mixed spiders build sheet webs underneath bryophyte mats Yule coalescent model, Pons et al. 2006; Fujisawa & Bar- on steep, north-facing rock outcrops (Coyle 2009). These raclough 2013) and arguably more powerful multilocus distinct outcrops are often separated by habitats lacking models. A plethora of multilocus species delimitation spiders (lower elevations, no spruce-fir forest, no rock methods have been developed over the past 10 years outcrops). Because of severe declines in spruce-fir forest (O’Meara 2010; Yang & Rannala 2010; Grummer et al. in the late 20th century, M. montivaga was formally 2014). Many empirical studies have been published listed as a US federally endangered species in 1995 (Fri- using such methods (reviewed in Fujita et al. 2012; Car- dell 1994, 2001). This status remains valid today, and stens et al. 2013), and this remains an active area of while recent survey work (Coyle 2009) has shown that method development (e.g. DISSECT, Jones et al. 2014; some montane populations include many rock outcrop unguided Bayesian Phylogenetics and Phylogeography demes and show comparatively large census population (BPP), Yang & Rannala 2014; *BFD, Leache et al. 2014). sizes (e.g. Great Smoky and Black Mountain popula- Despite this outstanding analytical progress, the tions), other populations are precariously small, known performance of multispecies coalescent methods in only from one or a few rock outcrops within a moun- highly genetically subdivided systems (e.g. taxa inhabit- tain range (e.g. Virginia Balsams, Plott Balsams). ing mountains, islands, caves) has not been extensively The research presented here was motivated primarily explored (Camargo & Sites 2013). A central assumption by an interest in documenting multilocus genetic struc- of many recently developed methods is the neutral coa- ture both within and among montane Microhexura pop- lescent (or something analogous), where gene trees ulations, with the goal of using this information to help evolve within species according to a no selection, no inform conservation decisions for this endangered recombination, panmixia model (Rannala & Yang 2003). species. There are no previous studies of Microhexura For example, the heuristic BROWNIE approaches devel- population structure, and while it is known that other oped by O’Meara (2010) simultaneously estimate spe- mygalomorphs (e.g. burrow-dwelling trapdoor spiders) cies trees and species limits by assuming unconstrained display remarkable microgeographic genetic differentia- gene flow within and a lack of gene flow between spe- tion (e.g. Bond & Stockman 2008; Hedin et al. 2013; cies. Simulations incorporating population structure Satler et al. 2013; Castalanelli et al. 2014; Opatova & tended to result in oversplitting by this method (O’Me- Arnedo 2014), whether such dispersal limitation applies ara 2010), and the empirical work of Niemiller et al. in the small-bodied (adult total length <6 mm), web- (2012) in naturally fragmented cavefishes hinted at building Microhexura is currently unknown. A second oversplitting, with both allelic and individual sampling goal was to explore the use of multilocus species delim- inflating species numbers. Another popular method is itation methods in this naturally fragmented system. BPP (Yang & Rannala 2010, 2014; Rannala & Yang 2013) The southern Appalachian mountains represent a hot- – this method assumes panmixia within species, and spot for speciation in both arthropods (e.g. Hedin 1997;

Fig. 1 Geographic distribution of sampled specimens from six primary montane populations. Number of sampled individuals and ‘demes’ (i.e. separate rock outcrop populations within mountain ranges) also indicated.

Thomas & Hedin 2008; Marek & Bond 2009; Hedin & sams. A subset of specimens was imaged using a Thomas 2010; Keith & Hedin 2012) and vertebrates (e.g. Visionary Digital BK Plus system (http://www.vision- Weisrock & Larson 2006; Crespi et al. 2010). Although arydigital.com), including a Canon 5D digital camera, Coyle (1981) did not comment on morphological geo- Infinity Optics Long Distance Microscope, P-51 camera graphic variation in his revision of M. montivaga,itis controller and FX2 lighting system. Individual images possible that morphologically cryptic speciation has were combined into a composite image using HELICON occurred in this taxon. Conversely, this system may FOCUS V5.3 and then edited using Adobe Photoshop represent a case where extreme population genetic CS6. structuring (but not speciation) leads to an oversplitting of lineage diversity by multispecies coalescent methods. Molecular marker development, sampling, data While these alternatives are challenging to resolve in collection any natural system, our results illustrate an empirical issue expected to become increasingly common as geno- Mitochondrial and nuclear genetic markers were devel- mic-scale data sets are gathered for taxa found in natu- oped specifically for Microhexura using comparative rally fragmented habitats. transcriptomic data (Material S1, Supporting information). Genomic DNA was extracted from nondestruc- tively sampled leg tissues, collected under permit by F. Materials and methods Coyle. Forty-seven M. montivaga specimens were sampled (Fig. 1, Material S2, Supporting information), plus Morphological study a specimen of the sister species M. idahoana from the The morphology of adult male pedipalps and first legs Pacific Northwest. Transcriptome data were used to (which possess mating spurs) is often used as a primary assess whether amplified PCR products were ‘on tar- character for species delimitation in mygalomorph spi- get’, but were not used in downstream nuclear analyses ders, including diplurids (e.g. Coyle 1984, 1988, 1995). as heterozygosity could not be assessed (transcriptomes To assess qualitative morphological divergence across were derived from multiple individuals, see Material M. montivaga populations, all adult males from Grand- S1, Supporting information). father Mountain (n = 2) and the Great Smoky Moun- After multiple iterations of primer testing and preli- tains (n = 8) were borrowed from the American minary sequencing, one mitochondrial and seven Museum of Natural History. Males are also known nuclear gene regions were chosen for comprehensive from the Black Mountains (vicinity Mt. Mitchell, includ- specimen sampling (Table 1; primers, PCR conditions ing type specimens – Crosby & Bishop 1925; Coyle and transcript annotations are provided in Material S3, 1981), but these specimens could not be located at the Supporting information). Amplified PCR products were AMNH. Adult males have never been collected from purified using standard techniques, and Sanger the Virginia Balsams, Roan Mountain or the Plott Bal- sequenced (29 coverage) at Macrogen USA. DNA

Table 1 Gene data summary information

Aligned Parsimony Primer Combo Length/No. Informative Nucleotide Name Sequences Substitution Model Clock Model Sites Diversity Exon or UTR mtDNA (COI) 1038/47 TN93 (partitioned) Strict 153 0.048 Exon B1_E3E4 664/62 HKY+I Strict 20 0.008 Exon, 30 UTR B1_C3C4 943/60 HKY+I Strict 10 0.004 50 UTR B1_D11D12 621/57 TN93 (partitioned) Strict 23 0.017 Exon B1_H3H4 432/64 TN93 (partitioned) Strict 10 0.010 Exon B2_F3F4 652/54 F81 Relaxed – Lognormal 9 0.006 No long ORFs, putative (*Strict used) noncoding B2_D3D4 783/54 GTR+I Strict 17 0.007 No long ORFs, putative noncoding B2_E7E8 582/58 HKY Strict 12 0.008 30 UTR

Parsimony informative sites and nucleotide diversity values calculated using MEGA 6.06 (Tamura et al. 2013), in-group data only. sequences were edited using Geneious Pro (www.gene- mer et al. 2014) and BPP (Yang & Rannala 2014) ‘valida- ious.com/) and trimmed to exclude primer sequences. tion’ analyses. Most validation analyses were based on Minor gaps in nuclear sequences were recoded as fol- nuclear-only data matrices, although we also conducted lows (*data matrices named using arbitrary primer BFD analyses using combined nuclear and mitochon- names): E3_E4 matrix, 17-base pair (bp) insertion found drial data. in Winter Star and Celo Knob, 6-bp deletion found in Smokies and Plott Balsams each recoded as two-state Discovery analyses nucleotide transitions (e.g. A<>G); D11D12 matrix, 3-bp insertion in out-group vs. in-group recoded as a two- Individual gene trees were estimated using maximum state nucleotide transition; B2_D3D4 matrix, 2 separate likelihood implemented in the RAxML_GUI (Stamatakis 2-bp insertions recoded as single two-state nucleotide 2006, 2014; Silvestro & Michalak 2012). Analyses transitions; B2_E7E8 matrix, single bp indel recoded as included a thorough bootstrap analysis (1000 bootstrap a two-state nucleotide transition. Heterozygous nuclear replicates) followed by multiple inferences (100) on sequences were bioinformatically phased to alleles alignments. Nuclear gene trees were estimated using an using the software program PHASE 2.1.1 (Stephens et al. unpartitioned GTR_Γ model; mitochondrial COI data 2001; Stephens & Donnelly 2003). SEQPHASE (Flot 2010) were partitioned by codon position, with the same was used to convert matrices for input into PHASE. PHASE model applied to each partition. analyses were conducted using default settings (phase The mitochondrial RAxML gene tree was used as threshold = 90%, 100 iterations, thinning interval = 1, input in bPTP analyses, implemented on the bPTP ser- burn-in = 100) and were repeated multiple times to ver (http://species.h-its.org/ptp/; Zhang et al. 2013). ensure consistent results. PTP is a single-locus species delimitation method using only nucleotide substitution information, implementing a model assuming gene tree branch lengths generated Analytical framework by two independent Poisson process classes (within- Our analytical framework included both ‘discovery’ and among-species substitution events). Available simu- and ‘validation’ approaches to delimit species (Ence & lation studies suggest that PTP outperforms GMYC Carstens 2011; Carstens & Satler 2013). Mitochondrial (Pons et al. 2006; Fujisawa & Barraclough 2013) for sin- gene trees and a mitochondrial Bayesian Poisson tree gle-locus species delimitations (Zhang et al. 2013). Two processes analysis (bPTP, Zhang et al. 2013) were used replicate bPTP analyses were run for 100 000 MCMC as species ‘discovery’ methods. Nuclear-only gene trees generations, with a thinning of 100 and burn-in of 0.1. and nuclear-only clustering results (POFAD, STRUCTURE) Multigenic nuclear genetic distances (uncorrected were similarly used as independent discovery methods. p-distances from PAUP*, Swofford 2002) among individ- These various genetic discovery results were combined uals were calculated using POFAD 1.05 (Joly & Bruneau with geographic criteria (e.g. isolated montane popula- 2006). To eliminate the potentially confounding inﬂu- tions as species) to formulate a set of alternative species ence of extreme female-based population structure, delimitation hypotheses, which were then statistically mitochondrial data were excluded in POFAD analyses. compared using Bayes factor delimitation (BFD; Grum- Also, out-group data were excluded from this analysis.

Analyses were conducted using both standardized sidered to identify an optimal K value – estimates from matrices (all individual matrices given the same weight) multiple replicates for multiple K values were calcu- and nonstandardized matrices (more variable matrices lated in Structure Harvester (Earl & vonHoldt 2012). with greater weight). Summary distances were used to Data were summarized using the FullSearch algorithm reconstruct NeighborNet networks in SplitsTree4 (Hu- of CLUMPP (Jakobsson & Rosenberg 2007) and visual- son & Bryant 2006). ized with DISTRUCT (Rosenberg 2004). STRUCTURE detects population structure through the use of allele frequencies, identifying genetically homo- Validation analyses geneous clusters of individuals that are in both Hardy– Weinberg and linkage equilibrium (Pritchard et al. 2000, Bayes factor delimitation (BFD; Grummer et al. 2014) is 2010). STRUCTURE has also been used to identify putative a recently developed approach that compares the mar- independently evolving genetic lineages in many stud- ginal likelihoods of competing species delimitation ies (e.g. Weisrock et al. 2010; Rittmeyer & Austin 2012; hypotheses using Bayes factors. Specifically, this Satler et al. 2013). Both mitochondrial and out-group method compares species tree models in which data were excluded in STRUCTURE analyses. SNAP Map sequences are assigned to differing numbers of lineages (Price & Carbone 2005; Aylor et al. 2006) and the Moby- (e.g. five species, six species) and chooses the model le SNAP workbench (Monacell & Cardone 2014) were that best explains the data (Grummer et al. 2014). Based used to convert DNA sequences to numbered unique on a combination of geography and ‘discovery’ genetic alleles (haplotypes) for STRUCTURE input (O’Neill et al. results (see above), putative Microhexura lineages were 2012). STRUCTURE runs were conducted assuming left separate or combined to generate eight alternative between 3 and 7 genetic clusters (K = 3 through K = 7), species delimitation hypotheses (Table 2). with each K value replicated three times. Analyses used A *BEAST species tree (inferred using *BEAST v1.8.0, an admixture model, with a burn-in of 1 9 105 steps Drummond et al. 2012) was estimated for each alterna- (with 1 9 106 MCMC steps after burn-in), and allele fre- tive hypothesis, using nuclear-only or nuclear plus quencies considered independent among populations. mitochondrial matrices, without out-groups. *BEAST Evanno et al. (2005) used simulations to show that the analyses were performed using 250 000 000 generations, maximal value of the log probability of the data given with data saved every 25 000 generations; the first 20% K (L(K)) does not necessarily provide an accurate esti- of each run was discarded as burn-in. For each hypoth- mation of K, but sometimes overestimates K. Instead, a esis, three to five *BEAST replicates were conducted to statistic called DK(= rate of change in log probability of ensure convergence and assessed using ESS values with data between successive K values) consistently provided TRACER v1.6 (Rambaut et al. 2014). Substitution models a more accurate estimate of K under the simulation for individual genes were chosen with JModeltest 2.1.6 conditions explored. Here, both approaches were con- (Guindon & Gascuel 2003; Darriba et al. 2012) using the

Table 2 Alternative species delimitation hypotheses tested using validation approaches

Hypothesis Distinct Species (total in parentheses) Motivation

H1 Plott Balsams, Smokies, Grandfather_Indian/Attic, Six geographic populations unique, two Grandfather_Watauga, Whitetop, Roan, Blacks, Blackstock_N (8) distinct genetic lineages at Grandfather Mountain and in Black Mountains

H2 Plott Balsams, Smokies, Grandfather, Whitetop, Roan, Six geographic populations unique, Blacks, Blackstock_N (7) two distinct lineages in Black Mountains

H3 (Plott Balsams + Smokies), Grandfather_Indian/Attic, Mitochondrial bPTP Grandfather_Watauga, Whitetop, Roan, Blacks, Blackstock_N (7)

H4 (Plott Balsams + Smokies), Grandfather, Whitetop, Roan, POFAD liberal Blacks, Blackstock_N (6) = H5 (Plott Balsams + Smokies), (Grandfather + Whitetop), STRUCTURE K 5 Roan, Blacks, Blackstock_N (5)

H6 (Plott Balsams + Smokies), Grandfather, Whitetop, Roan, Geographic populations unique except for (Blacks + Blackstock_N) (5) Plott Balsams + Smokies = H7 (Plott Balsams + Smokies), (Grandfather + Whitetop + Blackstock_N), STRUCTURE K 4 Roan, Blacks (4)

H8 (Plott Balsams + Smokies), Two species on opposite sides of (Grandfather + Whitetop + Roan + Blacks + Blackstock_N) (2) Asheville Basin

Bold values represent collapsed lineages from a preceding hypothesis.

AIC method (Table 1). All *BEAST analyses were con- was run for 50 000 generations, with results sampled ducted on the CIPRES Science Gateway (www.phy- every 5 generations; the first 1000 generations of each lo.org; Miller et al. 2010). Preliminary results suggested run were treated as burn-in. convergence issues when applying a relaxed clock model to the B2_F3F4 data, so a strict clock model was Results applied for all loci (Table 1). Marginal likelihoods were estimated using path-sampling (PS, Lartillot & Philippe Morphological divergence 2006) and stepping-stone (SS, Xie et al. 2011) methods, with 100 path steps, a chain length of 100 000 genera- We examined all male specimens from the AMNH, and tions and likelihoods saved every 100 generations. Mar- imaged a subset of these. Grandfather and Smokies ginal likelihood estimates (MLE) were averaged across male specimens are not obviously qualitatively different replicate runs to generate a single PS and SS value for in detail of the pedipalp and/or modified first leg each hypothesis. Bayes factors were then calculated by (Fig. 2, images submitted to Dryad). Although male taking the difference between the log of the best MLE specimens from the Black Mountains were not available and the log of other MLEs and multiplying each result for study, comparisons of imaged specimens to pub- * – by two [i.e. 2 (-lnHypA -lnHypB)]. The significance of lished drawings (Crosby & Bishop 1925 fig. 2; Coyle Bayes factor results was interpreted following Kass & 1981 fig. 15) do not indicate obvious differences. Raftery (1995), with 2lnBf >10 being considered as ‘decisive’ support for a hypothesis. Validation analyses were also performed using BPP v3.0 (Yang & Rannala 2010, 2014). This method utilizes a multispecies coalescent model and Bayesian statistics (A) (B) to delimit species. Traditionally, BPP analyses have relied on a user-specified guide tree to represent a species tree under a highly split delimitation hypothesis. A reversible-jump Markov chain Monte Carlo (rjMCMC) algorithm is then employed to estimate the posterior probability of different delimitation hypotheses by itera- tively collapsing or retaining nodes found in the guide tree (Yang & Rannala 2010). More recently, Yang & Rannala (2014) updated BPP to allow for joint inference of species limits and a species tree via a nearest neigh- bour interchange (NNI) algorithm that is able to signifi- cantly change the topology of the input guide tree. We implemented both the traditional (rjMCMC-only) and joint estimation methods (NNI + rjMCMC) using in-group-only nuclear matrices. (C) (D) For each BPP method, the nuclear-only *BEAST species tree corresponding to a liberal eight-species model

(H1, Table 2) was input as a guide tree. Three combinations of prior values specifying the ancestral population h s size ( ) and root age ( 0) were used, following Leache& Fujita (2010). These correspond to (i) large ancestral population sizes and deep divergences among species h ~ s ~ ( G(1,10) and 0 G(1,10)), (ii) small population sizes h ~ s ~ and shallow divergences ( G(2,2000) and 0 G (2,2000)) and (iii) large ancestral population sizes and h ~ s ~ shallow divergences ( G(1,10) and 0 G(2,2000)). A fourth combination representing small population sizes Fig. 2 Digital images of adult male morphology. Top structure in each panel is left pedipalp and bottom structure is left ﬁrst and intermediate divergence (h ~ G(2,2000) and s ~ G 0 leg, both in retrolateral view. (A, B) NC: Avery Co., Grandfa- (2,1000)) was also tested following Niemiller et al. ther Mountain, 17 November 1978, coll. F. Coyle, R. Bruce, J.D. (2012). For each set of prior combinations, replicate runs Pittillo; (C, D) NC: Swain Co., just below Clingman’s Dome were performed using different starting species trees, parking lot, Great Smoky Mountains National Park, 18 Octo- using both rjMCMC algorithms (0 and 1). Each analysis ber, 4 November 1978, coll. F. Coyle. All scale bars = 1 mm.

with microgeographic genetic differentiation along the Mitochondrial discovery analyses ridgeline of the Blacks (Material S4, Supporting infor- Maximum-likelihood analyses of mitochondrial data mation). A remarkable situation exists in the Blacks at (submitted to GenBank, Material S2, Supporting infor- Blackstock Knob, where highly divergent mitochondrial mation) show that different montane populations form clades (Blackstock_North vs. Blackstock_Northeast) divergent, well-supported (bootstrap values >70) clades exist on rock outcrops separated by just over 100 m. (Fig. 3). An exception includes the Plott Balsam speci- Sequences from Blackstock_Northeast fall into the Black mens, which are nested within the Smokies mitochon- Mountains mitochondrial clade, whereas sequences drial clade (Fig. 3). Also, mitochondrial data strongly from Blackstock_North form a separate divergent mito- support high divergence among demes within moun- chondrial clade (Fig. 3, Material S4, Supporting infor- tain ranges in several instances (Fig. 3). On Roan mation). Mountain, two geographically separate demes are The estimated number of mitochondrial bPTP species genetically distinct, and on Grandfather Mountain, two is between 7 and 25, with a mean value of 11 species. rock outcrop demes separated by approximately one The conservative number (7) corresponds exactly to kilometre (Indian House Cave_Attic Window vs. Wata- primary clades of the RAxML tree (Fig. 3) and implies uga View) form highly divergent mitochondrial clades. different species existing on Blackstock_North vs. Black- In the Black Mountains, almost all rock outcrop demes stock_Northeast rock outcrops (separated by ~ 100 m), form distinct mitochondrial microclades, consistent and two bPTP species on Grandfather Mountain (Indian

Whitetop 0.87 100 WHITETOP_MY4021 WHITETOP_MY4022 GRSM_MTLECONTE_MY1710 GRSM_MTLECONTE_MY1712 0.77 100 GRSM_MTLECONTE_MY1713 GRSM_MTCHAPMAN_MY1715 98 GRSM_MTCHAPMAN_MY1716 GRSM_MTLOVE_MY1709 Great Smokies GRSM_MTLOVE_MY1708 GRSM_MTLOVE_MY1707 Plott Balsams PLOTTBALSAMS_MY4023 96 PLOTTBALSAMS_MY4024 GRSM_MTCHAPMAN_MY1717 GRSM_MTBUCKLEY_MY1705 Grandfather GRSM_MTBUCKLEY_MY1704 GRSM_MTBUCKLEY_MY1706 GRSM_MTLECONTE_MY1711 0.95 100 GF_INDIANH_MY1342 GF_ATTIC_MY1338 ROANRESC_MY1353 0.63 100 ROANRESC_MY1354 Roan ROANHB_MY1352 ROANHB_MY1355 87 ROANHB_MY1351 97 BLACKS_CELOKNOB_MY1348 BLACKS_CELOKNOB_MY1347 96 BLACKS_MTMITCH2_MY4053 95 BLACKS_MTMITCH1_MY4052 BLACKS_GIBBES_MY4054 BLACKS_GIBBES_MY4055 BLACKS_GIBBES_MY4056 Blacks 70 BLACKS_POTATO_MY4058 BLACKS_POTATO_MY4057 BLACKS_BLACKSTOCKNE_MY4649 0.56 99 85 BLACKS_BLACKSTOCKNE_MY4659 98 BLACKS_BLACKSTOCKNE_MY4650 BLACKS_CATTAIL_MY1344 75 BLACKS_CATTAIL_MY1343 71 BLACKS_CATTAIL_MY1350 BLACKS_WINTERSTAR_MY1346 77 BLACKS_WINTERSTAR_MY1345 0.69 100 BLACKS_BLACKSTOCKN_MY4060 BLACKS_BLACKSTOCKN_MY4651 BLACKS_BLACKSTOCKN_MY4652 Blackstock_N Grandfather 79 GF_WATAUGAV_MY1339 GF_WATAUGAV_MY1341 0.95 100 GF_WATAUGAV_MY1340 0.02

Fig. 3 RA9ML mitochondrial gene tree. Bootstrap values >70 shown on branches; bPTP values in grey boxes. Gene tree arbitrarily rooted at longest internal branch. Colours used to designate montane populations (green = Great Smokies, black = Plott Balsams, dark orange = Whitetop, blue = Grandfather, purple = Roan Mtn., red = Blacks) used for most remaining ﬁgures; Blackstock _North deme designated with brown text and used for most remaining ﬁgures.

House Cave_Attic Window vs. Watauga View, demes the Evanno method. The maximal value of the log prob- separated by ~ 1 km). The bPTP posterior probability of ability of the data given K (L(K)) also peaks at K = 5 Plott Balsams separate from Great Smokies is zero. (546.9). Genetic clusters at K = 5 correspond to Smok- ies plus Plott Balsams, Roan Mountain, Black Moun- tains (except for Blackstock_North), Blackstock_North Nuclear discovery analyses and a single cluster that includes Grandfather Mountain Data were collected and phased for seven nuclear plus Whitetop Mountain (Fig. 5A). Although K = 5is regions (Material S2, Supporting information), with a optimal under the Evanno method, a more conservative total aligned length of 4677 bp (Table 1). Nuclear PCR- K = 4 hypothesis was also considered, with a K (L(K)) ampliﬁed Sanger data match transcriptome data (i.e. value (568.9) most similar to the preferred K value. PCR primers ampliﬁed the correct target gene region), The K = 4 result includes a single Grandfather + White- and nuclear matrices include minimal missing data (7 top + Blackstock_North lineage and suggests mixed nuclear gene matrices 9 47 in-group individuals per ancestry for many individuals (Fig. 5B). matrix – 3 total missing sequences, all in the B2_D3_D4 matrix, see Material S2, Supporting information). As Validation analyses expected, each nuclear gene tree is topologically unique (Material S5, Supporting information). Despite this Marginal likelihood values for alternative species delim- expected gene tree heterogeneity, there is clear phyloge- itation models assessed using BFD and combined netic signal in the nuclear data (Table 1), and several nuclear plus mitochondrial data are reported in Material generalizable patterns are apparent. All nuclear gene S6 (Supporting information). Marginal likelihoods trees include Plott Balsam alleles nested within a Smok- favour a seven-species model (H2), but are very similar ies clade, with Plott Balsam sequences identical to cer- to an alternative eight-species (H1) model (Material S6, tain Smokies sequences for all genes. The southwestern Supporting information). Both of these models imply Plott Balsam plus Smokies clade is distinct from north- different species on the Plott Balsams vs. the Smokies, eastern populations in most gene trees. Roan Mountain despite the fact that we found no evidence for either alleles form a clade in 4 of 7 gene trees, Whitetop mitochondrial or nuclear genetic divergence for these Mountain alleles form a clade in 4 of 7 gene trees, and allopatric populations (Figs 3–5). Because of these Grandfather Mountain alleles from a clade in 2 of 7 results, and following our general preference for conser- gene trees. The genetic relationship between Black- vative hypotheses, we hereafter emphasize the nuclear- stock_North vs. all other Black Mountains demes only validation results summarized below. (including Blackstock_Northeast) varies from gene to Bayes factor delimitation marginal likelihood values gene, but there is evidence for nuclear genetic for alternative species delimitation models assessed divergence of Blackstock_North in several gene trees. using nuclear-only data are shown in Table 3. The

Out-group sequences were only available for 3 nuclear favoured result is a seven-species model (H3) that corre- genes, and in-group root placement using out-group sponds to the conservative mitochondrial bPTP hypoth- information varies from gene to gene (Material S5, Sup- esis (Fig. 3), with internal species-level divergence both porting information). in the Black Mountains and on Grandfather Mountain, Nuclear POFAD results using standardized distances and a combined Plott Balsams plus Smokies lineage. generally correspond to those using nonstandardized However, this hypothesis does not differ from alterna- distances (Fig. 4). Because POFAD networks do not tive seven-species (H2) or six-species (H4) models at a include quantitative statistical criteria for delineating ‘decisive’ level of support (2lnBf <10, Table 3). The ‘distinct lineages’, deﬁning such units is somewhat arbi- more conservative six-species model includes separate trary. If geographic criteria are liberally used to recog- species on single mountaintops, but with two species in nize discrete genetic units (i.e. montane populations as the Blacks, and a single species in the Smokies plus discrete units), then all montane populations are dis- Plott Balsams. tinct except for the combined Plott Balsams plus Smok- Results of nuclear-only BPP analyses were largely ies lineage, which internally shows minimal congruent across runs, prior combinations and alterna- differentiation. Furthermore, there is POFAD evidence for tive (0,1) algorithms. Using the rjMCMC-only method, a nuclear separation between Blackstock_North vs. all most of the overall posterior probability is split between other Black Mountains demes, and the northernmost seven (H3)- and six (H4)-species models (Table 4), with demes in the Black Mountains (Celo Knob, Winter Star) higher posterior probabilities for the six-species model reveal evidence for limited nuclear differentiation. under small population size priors. Also, for the seven-

STRUCTURE analyses recover ﬁve genetic clusters, with species (H3) model, the nodal ‘speciation probability’ K = 5 including the largest DK (478) as estimated using for separate Grandfather species is always low

© 2015 John Wiley & Sons Ltd SKY ISLAND SPECIES DELIMITATION 3475 Great Smokies (A) Roan Plott Balsams ROAN_RESCLIFFS_MY1353 ROAN_ROANHB_MY1355 GRSMP_BUCKLEY_MY1705 Blacks PLOTTBALSAMS_MY4023 ROAN_ROANHB_MY1351 ROAN_ROANHB_MY1352 GRSMP_BUCKLEY_MY1706 GRSMP_MTLOVE_MY1709 GRSMP_MTLECONTE2_MY1713 GRSMP_MTLECONTE_MY1711 GRSMP_MTCHAPMAN_MY1716 GRSMP_MTCHAPMAN2_MY1717 BLACKS_WINTERSTAR_MY1345 PLOTTBALSAMS_MY4024 BLACKS_CELOKNOB_MY1347 BLACKS_WINTERSTAR_MY1346 ROAN_RESCLIFFS_MY1354 GRSMP_MTLECONTE_MY1710 BLACKS_CATTAIL_MY1343 GRSMP_MTLOVE_MY1707 GRSMP_MTLOVE_MY1708 BLACKS_CELOKNOB_MY1348 GRSMP_MTLECONTE2_MY1712 GRSMP_BUCKLEY_MY1704 GRSMP_MTCHAPMAN_MY1715 BLACKS_POTATO_MY4057 BLACKS_BLACKSTOCKNE_MY4649 BLACKS_MTMITCH1_MY4052 BLACKS_GIBBES_MY4054 BLACKS_POTATO_MY4058 BLACKS_CATTAIL_MY1350 BLACKS_MTMITCH2_MY4053 BLACKS_GIBBES_MY4056 BLACKS_GIBBES_MY4055 BLACKS_CATTAIL_MY1344 0.1

BLACKS_BLACKSTOCKNE_MY4650

BLACKS_BLACKSTOCKNE_MY4059 GRANDFATHER_WATAUGAV_MY1339 BLACKS_BLACKSTOCKN_MY4652 GRANDFATHER_WATAUGAV_MY1340 BLACKS_BLACKSTOCKN_MY4651 GRANDFATHER_ATTICWINDOW_MY1338 BLACKS_BLACKSTOCKN_MY4060 GRANDFATHER_INDIANHORSE_MY1342 Blackstock_North GRANDFATHER_WATAUGAV_MY1341 VA_WHITETOP_MY4022 Grandfather VA_WHITETOP_MY4021 Whitetop

GRSMP_MTCHAPMAN2_MY1717 GRSMP_MTLECONTE_MY1711 GRSMP_MTCHAPMAN_MY1715 (B) GRSMP_MTLECONTE2_MY1712 GRSMP_MTCHAPMAN_MY1716 Blackstock_North GRSMP_MTLECONTE2_MY1713 PLOTTBALSAMS_MY4024 GRSMP_MTLECONTE_MY1710 GRSMP_BUCKLEY_MY1706 GRSMP_MTLOVE_MY1709 PLOTTBALSAMS_MY4023 GRSMP_BUCKLEY_MY1704 BLACKS_BLACKSTOCKN_MY4060 GRSMP_BUCKLEY_MY1705 GRSMP_MTLOVE_MY1707 GRSMP_MTLOVE_MY1708 BLACKS_GIBBES_MY4056 BLACKS_BLACKSTOCKNE_MY4059 BLACKS_POTATO_MY4057 BLACKS_GIBBES_MY4055 BLACKS_CATTAIL_MY1350 BLACKS_BLACKSTOCKN_MY4651 BLACKS_MTMITCH2_MY4053 BLACKS_CATTAIL_MY1344 BLACKS_ BLACKS_BLACKSTOCKNE_MY4650 BLACKS_CATTAIL_MY1343 BLACKSTOCKN_MY4652 Great Smokies BLACKS_BLACKSTOCKNE_MY4649 BLACKS_MTMITCH1_MY4052 BLACKS_GIBBES_MY4054 0.001 BLACKS_POTATO_MY4058 Plott Balsams

BLACKS_CELOKNOB_MY1347 BLACKS_CELOKNOB_MY1348 BLACKS_WINTERSTAR_MY1345 BLACKS_WINTERSTAR_MY1346 Blacks ROAN_RESCLIFFS_MY1354 GRANDFATHER ROAN_ROANHB_MY1352 WATAUGAV_MY1339 ROAN_RESCLIFFS_MY1353 VA_WHITETOP_MY4021 GRANDFATHER_ ROAN_ROANHB_MY1355 Roan VA_WHITETOP_MY4022 INDIANHORSE_MY1342 ROAN_ROANHB_MY1351 GRANDFATHER_ATTICWINDOW_MY1338 GRANDFATHER_WATAUGAV_MY1341 Whitetop GRANDFATHER_WATAUGAV_MY1340 Grandfather

Fig. 4 NeighborNet networks reconstructed using POFAD distances derived from nuclear matrices. (A) Standardized distances. (B) Nonstandardized distances.

(PP < 0.60, Fig. 6A). Results using the NNI+rjMCMC is strongly supported. Other nodes in the northeastern method are generally congruent with those from the complex have low posterior probability values, indicat- rjMCMC-only method, with overall posterior probabili- ing that even though the northeastern genetic lineages ties split between seven-species (H3) and six-species are distinct, phylogenetic relationships among these lin- (H4) models, and higher posterior probabilities for the eages remain unclear (Fig. 6B). six-species model under small population size priors (Table 4). Discussion The nuclear-only *BEAST cloudogram estimated for a preferred six-species (H ) model (Fig. 6B) shows a pri- 4 Extreme population subdivision or cryptic speciation in mary well-supported split that separates the southwest- Microhexura? ern Smokies plus Plott Balsam clade from northeastern populations. Biogeographically, this primary division Two general patterns arise from consideration of revi- coincides with the low-elevation Asheville Basin barrier, sionary studies of other diplurid mygalomorphs (Coyle a conspicuous biogeographic barrier in the southern 1984, 1988, 1995). First, many of the species delimited Appalachians (e.g. Crespi et al. 2003, 2010; Weisrock & by Coyle exhibit intraspeciﬁc morphological variation, Larson 2006; Thomas & Hedin 2008). Within the north- and second, different described diplurid species are eastern complex, the Blacks plus Blackstock North node typically clearly different in male palpal and/or mating

(A) Fig. 5 STRUCTURE graphics (resulting from DISTRUCT) for (A) K = 5 and (B) K = 4. K = 5 Each column represents a specimen, grouped by deme of origin. Different colours represent different genetic clusters (K); estimated membership coefﬁcients Smokies + are proportional to bar colour height. Plott Balsams Blacks Roan Grandfather Whitetop Blackstock_North

l p e

ockN Cattail erStar MtLove RoanHB Whiteto RoanRes CeloKnob MtGibbes Wint MtMitchel MtBuckley MtLecont PotatoKnob IndianHorse Blackst MtChapman AtticWindow PlottBalsams WataugaView BlackstockNE MtLeconteNW

(B)

K = 4

Roan Smokies + Whitetop Blacks Plott Balsams Blackstock_North Grandfather

l p e ob bbes Cattail ockNE Gi MtLove RoanHB Whiteto atoKnob RoanRes tLecont CeloKn Mt WinterStar MtMitchel MtBuckley M Pot IndianHorse BlackstockN MtChapman AtticWindow PlottBalsams WataugaView Blackst MtLeconteNW spur morphology (Coyle 1984, 1988, 1995). We also note male mygalomorphs are naturally rare, and because that diplurid congeners are typically strictly allopatric M. montivaga is federally endangered. We did not sur- (e.g. within Allothele, Indothele, Andethele), which implies vey female morphology for two primary reasons – that allopatry per se does not promote genitalic stasis in female morphology is generally less taxonomically these spiders. When Coyle revised Microhexura, he informative in mygalomorphs, and in Microhexura spe- examined adults of both sexes from three montane pop- cifically, studies of Coyle (1981) indicate morphological ulations of M. montivaga, including Grandfather Moun- conservatism. In referring to M. montivaga vs. M. idaho- tain, the Black Mountains (vic. Mt. Mitchell) and the ana from the Pacific Northwest, Coyle stated that ‘if the Smokies (Coyle 1981). Coyle did not comment on either two ...species were not allopatric, it would be difficult male or female geographic morphological variation in to identify the females to species’. Our results point to a M. montivaga, but specifically discusses such variation need for the collection and study of males for all geneti- in the sister taxon M. idahoana (Coyle 1981; figs 21–43). cally distinct populations, and for the examination of Although our survey of male M. montivaga specimens is alternative character systems in both sexes (e.g. behav- consistent with the idea of morphological conservatism iour, chromosomes). However, we stress that the legal in this species, samples sizes are frustratingly small. status and general rarity of this taxon will make such This situation is challenging to overcome because adult studies challenging.

Table 3 Marginal likelihood and Bayes factor values for alter- from all sampled rock outcrops carry unique COI haplo- native species delimitation hypotheses, nuclear data only. Spe- types (Material S4, Supporting information). Evidence cies hypotheses as in Table 2 for genetic divergence also extends to the nuclear genome, and although we prefer a six-lineage hypothesis Stepping-Stone Path Sampling (Fig. 6), several analyses (e.g. BDF, BPP) recover seven ln (Marginal 2ln (Bayes ln (Marginal 2ln (Bayes distinct nuclear lineages that correspond exactly to seven Likelihood) Factor) Likelihood) Factor) divergent mitochondrial clades (Fig. 3). These congruent lineages are not strictly equivalent to montane popula- H1 (8 sp) 7921.95 17.94 7915.83 16.61 tions because two distinct genetic lineages exist on H2 (7 sp) 7917.76 9.57 7912.55 10.04 Grandfather Mountain and in the Black Mountains, and H3 (7 sp) 7912.98 N/A 7907.53 N/A the Plott Balsam population is not obviously genetically H4 (6 sp) 7916.60 7.24 7910.79 6.53 divergent from the Smokies population. H5 (5 sp) 7921.31 16.67 7915.99 16.92 H6 (5 sp) 7930.29 34.62 7925.11 35.17 Most genetic studies of mygalomorphs have been con- H7 (4 sp) 7952.03 78.10 7946.29 77.52 ducted on sedentary burrow-dwelling spiders (e.g. trap- H8 (2 sp) 8102.38 378.80 8093.81 372.57 door spiders) and reveal extensive population fragmentation (summarized in Bond et al. 2006; Satler Bold values represent collapsed lineages from a preceding et al. 2013). Our results for the small-bodied, web-build- hypothesis. ing Microhexura suggest that such fragmentation might also characterize mygalomorphs other than trapdoor spi- Patterns of genetic fragmentation in M. montivaga ders (see also Arnedo & Ferrandez 2007; Beavis et al. stand in stark contrast to strong morphological conserva- 2011; Castalanelli et al. 2014; Hamilton et al. 2014; Leavitt tism. Mitochondrial lineages are highly structured within et al. 2015). Additionally, while most prior genetic stud- and among montane populations (except for Smokies ies have relied upon mitochondrial data, our results sug- plus Plott Balsams), and northeast of the Asheville Basin, gest that strong population subdivision in both nuclear all mountaintops with multiple sampled rock outcrops and mitochondrial genomes may typify mygalomorphs show evidence for microgeographic genetic differentia- (see also Hedin et al. 2013; Satler et al. 2013; Leavitt et al. tion. For example, in the Black Mountains, specimens 2015). The modern availability of nuclear genomic

Table 4 Results of Bayesian Phylogenetics and Phylogeography analyses. Posterior probability values for six-species (H4) and seven- species (H2 and H3) models shown. Six-species model preferred in all cases except those bolded

# In-group Species Delimited (PPa)

Priors Run #, Algorithm Used

hs 0 1,0 2,0 1,1 2,1 rjMCMC-only method

Large pop, shallow div G (1,10) G (2,2000) H3 (0.54) H3 (0.47) H3 (0.47) H3 (0.36) H4 (0.42) H4 (0.50) H4 (0.53) H4 (0.60) Small pop, shallow div G (2,2000) G (2,2000) H3 (0.26) H3 (0.24) H3 (0.24) H3 (0.23) H4 (0.70) H4 (0.73) H4 (0.74) H4 (0.75) Large pop, deep div G (1,10) G (1,10) H3 (0.35) H3 (0.45) H3 (0.42) H3 (0.40) H4 (0.47) H4 (0.55) H4 (0.58) H4 (0.60) Small pop, inter div G (2,2000) G (2,1000) H3 (0.22) H3 (0.22) H3 (0.23) H3 (0.22) H4 (0.75) H4 (0.74) H4 (0.76) H4 (0.76) NNI+rjMCMC method

Large pop, shallow div G (1,10) G (2,2000) H2 (0.11) H2 (0.07) H3 (0.40) H3 (0.41) H3 (0.35) H3 (0.30) H4 (0.38) H4 (0.57) H4 (0.65) H4 (0.50) Small pop, shallow div G (2,2000) G (2,2000) H3 (0.24) H3 (0.21) H3 (0.22) H3 (0.22) H4 (0.73) H4 (0.75) H4 (0.74) H4 (0.75) Large pop, deep div G (1,10) G (1,10) H2 (0.27) H3 (0.34) H3 (0.36) H3 (0.22) H3 (0.50) H4 (0.61) H4 (0.63) H4 (0.32) H4 (0.49) Small pop, inter div G (2,2000) G (2,1000) H3 (0.18) H3 (0.24) H3 (0.19) H3 (0.19) H4 (0.80) H4 (0.73) H4 (0.78) H4 (0.79)

(A) nuc BFD conservative Validation Fig. 6 (A) Bayesian Phylogenetics and Phylogeography guide tree with posterior probability values for species delimi- nuc STR K = 4 tations, summarized across runs, priors nuc STR and algorithms (0, 1). For the rjMCMC- K = 5 only method, these values correspond to nPOFAD ‘speciation probabilities’ at nodes (values Discovery + mtDNA NOT italicized). For the NNI rjMCMC bPTP method (italicized values), posterior mtDNA probabilities are reported for the pres- gene tree ence of single distinct tip lineages (e.g. PP values for Plott Balsams plus Smokies Plott GF Wat collapsed as single lineage range from Smokies Balsams GF Indian BStock_N Roan * Blacks _Attic 0.55 to 1.0). (B) Nuclear-only BEAST spe- 0.55–1.0 1.0 1.0 0.49–0.81 1 . 0 Whitetop1.0 cies tree estimated for hypothesis four (six species), with posterior probability 0.22–0.56 0–0.19 1.0 values at nodes. The posterior distribution of species trees was visualized using 1.0 DENSITREE v2.0.1 (Bouckaert 2010); the ‘root canal’ option was chosen to high- 1.0 light the topology of the MCC tree (in blue), and the next two most frequent topologies are drawn with red and green 1.0 branches, respectively.

1.0

(B) SMOK_PB GF ROAN WT BLACKS BLACKS_N 1.0 0.69 0.48 0.89 1.0 resources for nonmodel taxa will allow this hypothesis to specimens associated with bryophyte mats on, or at the be more broadly tested in the coming years. base of, rock outcrops in high-elevation spruce-fir Bernardo (2011) defined cryptic species as ‘highly forest. Searches in other microhabitats or substrates, evolutionary divergent lineages that share a very simi- even in appropriate forests, did not reveal spiders. Ulti- lar morphology and which are therefore difficult to dis- mately, the biology of cryptic species narrows the types tinguish based upon morphological discontinuities’. of data available to investigate species boundaries. It is Appalachian Microhexura might include cryptic species, for these exact reasons that genetic data have been so and such a finding in a mygalomorph taxon would cer- valued as a means to discover and objectively delimit tainly not be unexpected (e.g. Bond et al. 2001; Cooper cryptic species (Zhang et al. 2011; Fujita et al. 2012; et al. 2011; Hedin et al. 2013; Satler et al. 2013; Castala- Niemiller et al. 2012). nelli et al. 2014; Opatova & Arnedo 2014; Leavitt et al. 2015). Cryptic species often share biological features Multispecies coalescent and the challenge of naturally beyond morphological conservatism. For example, fragmented systems microhabitat specialization is also common, with stabi- lizing selection thus promoting both ecological niche Multispecies coalescent methods for genetic species and morphological conservatism (reviewed in Keith & delimitation have been highlighted as statistically rigor- Hedin 2012). Microhabitat specialization clearly applies ous, objective and repeatable (Fujita et al. 2012; Camargo in M. montivaga, as Coyle (2009) found essentially all & Sites 2013; Carstens et al. 2013). However, multispecies

© 2015 John Wiley & Sons Ltd SKY ISLAND SPECIES DELIMITATION 3479 coalescent methods (and genetics-only methods more 2013) and plants (Carstens & Satler 2013). These sug- generally) can fail under certain speciation scenarios, for gestions follow from comparison of BPP results to example if speciation is very recent, or involves strong other lines of evidence and with results from other selection in a small fraction of the genome, with gene genetic analyses. False-positive rates have also been flow over the rest of the genome (Fujita et al. 2012; assessed via simulation for BFD (Grummer et al. 2014), Sousa & Hey 2013). These are largely ‘too much gene and although oversplitting rates were found to be neg- flow’ scenarios. As such, authors have recently devel- ligible, these simulations did not explicitly include oped methods that formally integrate multiple lines of internal population structure as in Zhang et al. (2011). evidence in a multispecies coalescent framework (e.g. Because BFD is founded on the multispecies coalescent iBPP, Solıs-Lemus et al. 2015), consistent with principles model of *BEAST, which again assumes the neutral coa- of integrative taxonomy (Dayrat 2005; Schlick-Steiner lescent, poor model fit resulting from population struc- et al. 2010; Derkarabetian & Hedin 2014). However, ture (see Reid et al. 2014) could potentially impact BFD because of the ‘data narrowing’ problem discussed results. above, speciation biologists working on cryptic species While it remains unclear how many species exist in often are forced to rely on genetics-only delimitations. the Microhexura montivaga ‘complex’, this taxon clearly If such delimitations are biased to mistake population illustrates an empirical issue that requires additional structure for species-level divergence, then ‘too little methodological attention. The multispecies coalescent gene flow’ also becomes a problem area for multispe- BPP method has been claimed to delimit biological species coalescent methods (O’Meara 2010; Niemiller et al. cies, but Zhang et al. (2011) acknowledge that ‘the 2012; Barley et al. 2013; McKay et al. 2013). method does not take into account whether the ... low Hey (2009) argued that many genetic species delimi- migration rate is due to geographical barriers or to tation approaches are biased towards oversplitting intrinsic reproductive isolation’ and that ‘two allopatric because they confound population structure with speci- populations that diverge due to neutral drift without ation. Evidence for this can be seen in the single-locus establishment of reproductive barriers may be inferred to results presented here (e.g. mean value of 11 bPTP be two species...’. Of course, this is not a problem with species), and other studies have convincingly shown the method per se, as the method effectively recognizes oversplitting in single-locus GMYC analyses (e.g. Lohse a low-geneflow threshold (i.e. M = Nm 1). In many 2009; Keith & Hedin 2012; Miralles & Vences 2013; empirical systems, measuring highly reduced gene flow, Satler et al. 2013). Multispecies coalescent methods are in the light of additional data, would indeed provide not immune to oversplitting, for reasons well summa- strong evidence for species status. However, in frag- rized by Niemiller et al. (2012). Low gene flow impacts mented systems with naturally constrained gene flow, the coalescent process, as genetic lineages sampled isolated populations and species are potentially equiva- from the same geographic location are more likely to lent under this model. Naturally fragmented systems coalesce locally than with lineages from other locations, with viscous gene flow are common in nature, and resulting in more congruence across independent gene modern access to genomic-scale data sets makes popu- trees than is expected under the neutral coalescent lation differentiation ever easier to measure. Unless (Kuo & Avise 2005). To assess the possibility of false biologists are willing to recognize all genetically diver- positives (oversplitting), extensive simulations have gent populations as species, there is a clear need to been conducted for guided BPP (Zhang et al. 2011). allow and incorporate population structure as a param- The most relevant simulations included a linear step- eter in multispecies coalescent methods (Camargo et al. ping-stone model with four populations of equal size, 2012; Niemiller et al. 2012; Camargo & Sites 2013; with symmetrical migration at variable levels (number Carstens et al. 2013; Leache et al. 2014). of migrants per generation M = 0.001–100) constrained to occur among adjacent populations. Under these sim- Regional biogeography ulation conditions, two species were inferred (at posterior probability >0.80) only when M values were <1 Both paleobotanical and genetic data indicate that the (using 10 loci). Because the authors essentially equate current sky island distribution of spruce-fir forests in the such low M values with ‘speciation’ (see below), the southern Appalachians is an evolutionarily recent phe- authors claim that ‘Bayesian inference may be quite nomenon (Delcourt & Delcourt 1984; Potter et al. 2008). robust to complex population structures’. Alternatively, At the last glacial maximum (LGM) and at multiple gla- BPP has been suggested to potentially oversplit in cial maxima earlier in the Pleistocene, these forests complex, low-geneflow empirical systems, including existed more expansively and continuously at lower ele- trapdoor spiders (Satler et al. 2013), birds (McKay et al. vations (up to 1000 metres lower), with mostly treeless (2013), lizards (Barley et al. 2013; Miralles & Vences tundra at the highest elevations (Delcourt & Delcourt

1984). Only recently (8000–4000 years ago) have these species M. montivaga is actually a complex of mostly forests become restricted to the highest southern Appala- allopatric, highly distinctive genetic lineages (Fig. 6). chian peaks. Animals restricted to southern Appalachian Most of these lineages are microendemic, restricted to a sky islands, presuming ecological niche conservation single mountain range or even a single rock outcrop. through time, are expected to show similar ‘recent frag- Even if these lineages do not constitute cryptic species, mentation’ biogeographic patterns. Although many mod- the level of evolutionary independence measured makes ern studies have considered regional biogeography (e.g. them clear candidates for lineage-specific conservation Weisrock & Larson 2006; Thomas & Hedin 2008; Hedin & decisions (i.e. as ‘management units’ or ‘evolutionary Thomas 2010; Keith & Hedin 2012), few have focused on significant units’). Conservation recommendations for high-elevation spruce-fir specialists. Two notable animal the preferred six-lineage hypothesis are made below. exceptions include pygmy salamanders (Desmognathus Virginia Balsams (Whitetop) – This geographically wrighti, Crespi et al. 2003, 2010) and smoky shrews (Sorex isolated population carries unique genetic diversity in fumeus, Sipe & Browne 2004). In Sorex fumeus, mitochon- both mitochondrial and nuclear genomes (Fig. 6). Coyle drial data indicate reduced gene flow across the low-ele- (2009) estimated 13 hectares of suitable habitat on vation Asheville Basin barrier and between individual Whitetop Mountain (in pure Spruce forest) and found northeastern montane populations (e.g. Blacks, Roan very low relative spider abundances. Pine Mountain, Mtn., Grandfather Mtn., Whitetop Mtn.). Although for- also part of the Virginia Balsams, includes a spider pop- mal dating analyses were not conducted, Sipe & Browne ulation that was not sampled in this study. Continued (2004) hypothesized a post-LGM fragmentation model close monitoring is needed at both locations, as loss of for this taxon. the Virginia Balsams population would constitute a con- Genetic patterns in Desmognathus are geographically siderable loss of unique genetic diversity. Grandfather similar to both Sorex and Microhexura, with populations Mountain – Multiple rock outcrop demes exist on on opposite sides of the Asheville Basin, and genetically Grandfather Mountain (Coyle 2009). The two demes distinct northeastern populations restricted to the Black, sampled constitute divergent mitochondrial lineages Roan, Grandfather and Virginia Balsam mountains and show weak evidence for nuclear differentiation in (Crespi et al. 2003). In this taxon, independent molecu- BFD and BPP analyses (Fig. 6). Sampling of additional lar clock analyses of both mitochondrial and nuclear al- rock outcrop demes might uncover additional genetic lozyme data suggest Pliocene ages for the primary lineages on Grandfather Mountain, and management to southwest–northeast Asheville Basin divergence, and maintain all genetically diverse demes within this range more recent studies of ecology, morphology and genet- should be a priority. Roan Mountain – This allopatric ics support the elevation of a distinct species (D. organi) population carries unique genetic diversity in both northeast of the Asheville Basin (Crespi et al. 2010). mitochondrial and nuclear genomes (Fig. 6). Multiple Genetic data indicate that D. organi populations are rock outcrop demes exist on Roan Mountain (Coyle more isolated and smaller in size compared to south- 2009); the two sampled demes differ slightly for mito- western D. wrighti populations, and although diver- chondrial genes, but are similar in their nuclear gence time estimates for these populations include wide genomes. Black Mountains – Multiple rock outcrop confidence intervals, these estimates appear inconsistent demes exist in the Blacks (Coyle 2009). Most of these with a post-LGM fragmentation model (mean of 1 mil- demes carry unique mitochondrial lineages, consistent lion years; Crespi et al. 2003). We hypothesize that with microgeographic genetic differentiation along the Microhexura divergence times follow a Pliocene/Early ridgeline of the Blacks (Material S4, Supporting infor- Pleistocene model as found in pygmy salamanders. An mation). Management to maintain all genetically diverse intriguing implication of this model is that Microhexura demes within this range should be a priority. Black- populations lack complete historical niche conservation, stock_North – This distinct genetic lineage (Fig. 6) is with possible recent (e.g. LGM) habitat occupation at not strictly geographically isolated, but instead occupies elevations above spruce-fir forests, despite the fact that a single rock outcrop at the southwestern edge of the all modern populations are restricted to these forests Black Mountains system (Figs 1 and 3). Because the (Coyle 2009). We intend to explore these hypotheses Blackstock_North genetic lineage is only found on a sin- further in a separate manuscript. gle rock outcrop, this deme should be monitored very closely, and additional rock outcrops in the vicinity should be sampled for this lineage. Again, loss of this Conservation recommendations for a federally single deme would constitute a considerable loss of endangered species complex unique genetic diversity. Great Smokies plus Plott The data presented here have important conservation Balsams – This distinct genetic lineage (Fig. 6) occurs implications, showing that the federally endangered southwest of the low-elevation Asheville Basin barrier

© 2015 John Wiley & Sons Ltd SKY ISLAND SPECIES DELIMITATION 3481 and is therefore arguably the most geographically trapdoor spider Aptostichus simus. Molecular Ecology, 10, 899– isolated of all lineages. Multiple rock outcrop demes 910. exist in the Smokies (Coyle 2009), but the sampled de- Bond JE, Beamer DA, Lamb T, Hedin MC (2006) Combining genetic and geospatial analyses to infer population extinction mes are relatively genetically homogeneous. The geo- in mygalomorph spiders endemic to the Los Angeles region. graphically isolated Plott Balsams population is Animal Conservation, 9, 145–157. essentially genetically identical to the Smokies popula- Bouckaert RR (2010) DensiTree: making sense of sets of phylo- tion. From a genetics perspective, translocation of speci- genetic trees. Bioinformatics, 26, 1372–1373. mens from the Great Smokies to the Plott Balsams Camargo A, Sites J Jr (2013) Species delimitation: a decade after would be justified, which may be required given the the renaissance. In: The Species Problem - Ongoing Issues (ed. – very low relative spider abundances and limited habitat Pavlinov I), pp. 225 247. InTech, New York. Camargo A, Morando M, Avila LJ, Sites JW Jr (2012) Species (estimated 3 hectares) in the Plott Balsams (Coyle 2009). delimitation with ABC and other coalescent-based methods in lizards of the Liolaemus darwinii complex (Squamata: Liola- Acknowledgements emidae). Evolution, 66, 2834–2849. Carstens BC, Satler JD (2013) The carnivorous plant described Sue Cameron has provided tremendous support for this project as Sarracenia alata contains two cryptic species. Biological Jour- and deserves special thanks. Sue Cameron, John Fridell and Bill nal of the Linnean Society, 4, 737–746. Commins assisted in the permitting process. Jason Bond, Bill Carstens BC, Pelletier TA, Reid NM, Satler JD (2013) How to Shear and Sue Cameron helped to collect specimens. Kristen fail at species delimitation. Molecular Ecology, 22, 4369–4383. Emata provided laboratory assistance, with help from David Castalanelli MA, Teale R, Rix MG, Kennington WJ, Harvey MS Zezoff; James Starrett and Nicole Garrison extracted RNA; Sha- (2014) Barcoding of mygalomorph spiders (Araneae: Mygalo- han Derkarabetian helped in the transcriptome assembly pro- morphae) in the Pilbara bioregion of Western Australia cess. Isabella Niewiadomski helped to image specimens. Lou reveals a highly diverse biota. Invertebrate Systematics, 28, Sorkin and Lorenzo Prendini facilitated the specimen loan from 375–385. the AMNH. Comments from Shahan Derkarabetian, Jared Cooper SJB, Harvey MS, Saint KM, Main BY (2011) Deep phy- Grummer, Dean Leavitt, Jordan Satler and Jim Starrett logeographic structuring of populations of the trapdoor spi- improved the manuscript. This research was funded by the US der Moggridgea tingle (Migidae) from southwestern Australia: Fish & Wildlife Service, contract #F12AP00739. evidence for long-term refugia within refugia. Molecular Ecol- ogy, 20, 3219–3236. Microhexura References Coyle FA (1981) The mygalomorph spider genus (Araneae, Dipluridae). Bulletin American Museum of Natural – Arnedo MA, Ferrandez M (2007) Mitochondrial markers reveal History, 170,64 75. deep population subdivision in the European protected spi- Coyle FA (1984) A revision of the African mygalomorph spider der Macrothele calpeiana (Walckenaer, 1805) (Araneae, Hexa- genus Allothele (Araneae, Dipluridae). American Museum Nov- – thelidae). Conservation Genetics, 8, 1147–1162. itates, 2794,1 20. Aylor DL, Price EW, Carbone I (2006) SNAP: Combine and Coyle FA (1985) Observations on the mating behaviour of the Map modules for multilocus population genetic analysis. Bio- tiny mygalomorph spider, Microhexura montivaga Crosby & informatics, 22, 1399–1401. Bishop (Araneae, Dipluridae). Bulletin British Arachnological – Barley AJ, White J, Diesmos AC, Brown RM (2013) The chal- Society, 6, 328 330. lenge of species delimitation at the extremes: Diversification Coyle FA (1988) A revision of the American funnel-web my- without morphological change in Philippine sun skinks. Evo- galomorph spider genus Euagrus (Araneae, Dipluridae). Bul- – lution, 67, 3556–3572. letin American Museum of Natural History, 187, 203 292. Beavis AS, Sunnucks P, Rowell DM (2011) Microhabitat prefer- Coyle FA (1995) A revision of the funnelweb mygalomorph ences drive phylogeographic disparities in tow Australian spider subfamily Ischnothelinae (Araneae, Dipluridae). Bulle- – funnel web spiders. Biological Journal of the Linnean Society, tin American Museum of Natural History, 226,1 133. 104, 805–819. Coyle FA (2009) Status survey of the spruce-fir moss spider, Mi- Bernardo J (2011) A critical appraisal of the meaning and diag- crohexura montivaga. Report to the US Fish & Wildlife Service. nosability of cryptic evolutionary diversity, and its implica- Crespi EJ, Rissler LJ, Browne RA (2003) Testing Pleistocene tions for conservation in the face of climate change. In: refugia theory: phylogeographical analysis of Desmognathus Climate Change, Ecology and Systematics (eds Hodkinson TR, wright, a high-elevation salamander in the southern Appala- – Jones MB, Waldren S, Parnell JAN), pp. 380–438. Cambridge chians. Molecular Ecology, 12, 969 984. University Press, Cambridge, UK. Crespi EJ, Browne RA, Rissler LJ (2010) Taxonomic revision of Bond JE, Stockman AK (2008) An integrative method for Desmognathus wrighti (Caudata: Plethodontidae). Herpetolog- – delimiting cohesion species: Finding the population-species ica, 66, 283 295. interface in a group of Californian trapdoor spiders with Crosby CR, Bishop SC (1925) Two new spiders from the Blue extreme genetic divergence and geographic structuring. Sys- Ridge mountains of North Carolina (Araneina). Entomological – tematic Biology, 57, 628–646. News, 36, 142 146. Bond JE, Hedin MC, Ramirez MG, Opell BD (2001) Deep Darriba D, Taboada GL, Doallo R, Posada D (2012) jModelTest molecular divergence in the absence of morphological and 2: more models, new heuristics and parallel computing. Nat- ecological change in the Californian coastal dune endemic ure Methods, 9, 772.

Dayrat B (2005) Towards integrative taxonomy. Biological Jour- Theraphosidae). Molecular Phylogenetics and Evolution, 71,79– nal of the Linnean Society, 85, 407–415. 93. Delcourt H, Delcourt P (1984) Late quaternary history of the Hedin M (1997) Speciational history in a diverse clade of spruce-fir ecosystem in the southern Appalachian Mountain habitat-specialized spiders (Araneae: Nesticidae: Nesticus): region. In: The Southern Appalachian Spruce-Fir Ecosystem: Its inferences from geographic-based sampling. Evolution, 51, Biology and Threats (ed. White PS), pp. 22–35. National Park 1929–1945. Service, Southeast Region, Research/Resource Management Hedin M, Thomas SM (2010) Molecular systematics of east- Report SER-71, Gatlinburg, Tennessee, USA. ern North American Phalangodidae (Arachnida: Opiliones: Derkarabetian S, Hedin M (2014) Integrative taxonomy and Laniatores), demonstrating convergent morphological evo- species delimitation in harvestmen: a revision of the western lution in caves. Molecular Phylogenetics and Evolution, 54, North American genus Sclerobunus (Opiliones: Laniatores: 107–121. Travunioidea). PLoS ONE, 9, e104982. Hedin M, Starrett J, Hayashi C (2013) Crossing the uncrossable: Derkarabetian S, Ledford J, Hedin M (2011) Genetic diversifica- novel trans-valley biogeographic patterns revealed in the tion without obvious genitalic morphological divergence in genetic history of low-dispersal mygalomorph spiders (An- harvestmen (Opiliones, Laniatores, Sclerobunus robustus) from trodiaetidae, Antrodiaetus) from California. Molecular Ecology, montane sky islands of western North America. Molecular 22, 508–526. Phylogenetics and Evolution, 61, 844–853. Hey J (2009) On the arbitrary identification of real species. In: Drummond AJ, Suchard MA, Xie D, Rambaut A (2012) Bayes- Speciation and Patterns of Diversity (eds Butlin RK, Bridle JR, ian phylogenetics with BEAUti and the BEAST 1.7. Molecular Schluter D), pp. 15–28. Cambridge University Press, Cam- Biology & Evolution, 29, 1969–1973. bridge, UK. Earl DA, vonHoldt BM (2012) STRUCTURE HARVESTER: a Hey J, Pinho C (2012) Population genetics and objectivity in website and program for visualizing STRUCTURE output species diagnosis. Evolution, 66, 1413–1429. and implementing the Evanno method. Conservation Genetic Huson DH, Bryant D (2006) Application of phylogenetic net- Resources, 4, 359–361. works in evolutionary studies. Molecular Biology & Evolution, Ence DD, Carstens BC (2011) SpedeSTEM: a rapid and accurate 23, 254–267. method for species delimitation. Molecular Ecology Resources, Jakobsson M, Rosenberg NA (2007) CLUMPP: a cluster match- 11, 473–480. ing and permutation program for dealing with label switch- Evanno G, Regnaut S, Goudet J (2005) Detecting the number of ing and multimodality in analysis of population structure. clusters of individuals using the software STRUCTURE: a Bioinformatics, 23, 1801–1806. simulation study. Molecular Ecology, 14, 2611–2620. Joly S, Bruneau A (2006) Incorporating allelic variation for Flot JF (2010) SeqPHASE: a web tool for interconverting reconstructing the evolutionary history of organisms from PHASE input/output files and FASTA sequence alignments. multiple genes: an example from Rosa in North America. Molecular Ecology Resources, 10, 162–166. Systematic Biology, 55, 623–636. Fridell JA (1994) Endangered and threatened wildlife and Jones G, Aydin Z, Oxelman B (2014) DISSECT: an assignment- plants; proposal to list the spruce-fir moss spider as an free Bayesian discovery method for species delimitation endangered species. Federal Register, 59, 3825–3829. under the multispecies coalescent. Bioinformatics, 31, 991–998. Fridell JA (2001) Endangered and threatened wildlife and Kass RE, Raftery AE (1995) Bayes factors. Journal of the Ameri- plants; reopening of public comment period and notice of can Statistical Association, 90, 773–795. availability of draft economic analysis for proposed critical Keith RM, Hedin M (2012) Extreme mitochondrial population habitat determination for the spruce-fir moss spider. Federal subdivision in southern Appalachian paleoendemic spiders Register, 66, 9806–9808. (Araneae, Hypochilidae, Hypochilus), with implications for Fujisawa T, Barraclough TG (2013) Delimiting species using species delimitation. Journal of Arachnology, 40, 167–181. single-locus data and the Generalized Mixed Yule Coalescent Kuo CH, Avise JC (2005) Phylogeographic breaks in low-dis- approach: a revised method and evaluation on simulated persal species: the emergence of concordance across gene data sets. Systematic Biology, 62, 707–724. trees. Genetica, 124, 179–186. Fujita MK, Leache AD, Burbrink FT, McGuire JA, Moritz C Lartillot N, Philippe H (2006) Computing Bayes factors using (2012) Coalescent-based species delimitation in an integrative thermodynamic integration. Systematic Biology, 55, 195–207. taxonomy. Trends in Ecology & Evolution, 27, 480–488. Leache AD, Fujita MK (2010) Bayesian species delimitation in Grummer JA, Bryson RW Jr, Reeder TW (2014) Species delimi- West African forest geckos (Hemidactylus fasciatus). Proceed- tation using Bayes factors: simulations and application to the ings of the Royal Society B: Biological Sciences, 277, 3071–3077. Sceloporus scalaris species group (Squamata: Phrynosomati- Leache AD, Fujita MK, Minin VN, Bouckaert RR (2014) Species dae). Systematic Biology, 63, 119–133. delimitation using genome-wide SNP data. Systematic Biol- Guindon S, Gascuel O (2003) A simple, fast and accurate ogy, 63, 534–542. method to estimate large phylogenies by maximum-likeli- Leavitt DH, Bezy RL, Crandall KA, Sites JW Jr (2007) Multi- hood”. Systematic Biology, 52, 696–704. locus DNA sequence data reveal a history of deep cryptic Hamilton CA, Hendrixson BE, Brewer MS, Bond JE (2014) vicariance and habitat-driven convergence in the desert night An evaluation of sampling effects on multiple DNA lizard Xantusia vigilis species complex (Squamata: Xantusii- barcoding methods leads to an integrative approach for dae). Molecular Ecology, 16, 4455–4481. delimiting species: a case study of the North American Leavitt DH, Starrett J, Westphal M, Hedin M (2015) Multilocus tarantula genus Aphonopelma (Araneae, Mygalomorphae, sequence data reveal dozens of putative cryptic species in a

radiation of endemic Californian mygalomorph spiders (Ara- Rannala B, Yang Z (2003) Bayes estimation of species diver- neae, Mygalomorphae, Nemesiidae). Molecular Phylogenetics gence times and ancestral population sizes using DNA & Evolution, doi: 10.1016/j.ympev.2015.05.016. sequences from multiple loci. Genetics, 164, 1645–1656. Lohse K (2009) Can mtDNA barcodes be used to delimit spe- Rannala B, Yang Z (2013) Improved reversible jump algorithms cies? A response to Pons et al. (2006). Systematic Biology, 58, for Bayesian species delimitation. Genetics, 194, 245–253. 439–442. Reid NM, Brown JM, Satler JD et al. (2014) Poor fit to the Maddison WP, McMahon MM (2000) Divergence and reticu- multi-species coalescent model is widely detectable in empir- lation among montane populations of the jumping spider ical data. Systematic Biology, 63, 322–333. Habronattus pugillis Griswold. Systematic Biology, 49, 400– Rittmeyer EN, Austin CC (2012) The effects of sampling on 421. delimiting species from multi-locus sequence data. Molecular Marek PE, Bond JE (2009) A Mullerian€ mimicry ring in Appa- Phylogenetics & Evolution, 65, 451–463. lachian millipedes. Proceedings of the National Academy of Sci- Rosenberg NA (2004) Distruct: a program for the graphical display ences, USA, 106, 9755–9760. of population structure. Molecular Ecology Notes, 4, 137–138. McKay BD, Mays HL Jr, Wu Y et al. (2013) An empirical com- Rovito SM (2010) Lineage divergence and speciation in the parison of character-based and coalescent-based approaches web-toed salamanders (Plethodontidae: Hydromantes) of the to species delimitation in a young avian complex. Molecular Sierra Nevada, California. Molecular Ecology, 19, 4554–4571. Ecology, 22, 4943–4957. Satler JD, Carstens BC, Hedin M (2013) Multilocus species Miller MA, Pfeiffer W, Schwartz T (2010). Creating the CIPRES delimitation in a complex of morphologically conserved trap- science gateway for inference of large phylogenetic trees. door spiders (Mygalomorphae, Antrodiaetidae, Aliatypus). Proceedings of the Gateway Computing Environments Systematic Biology, 62, 805–823. Workshop (GCE), New Orleans, 2010, 1-8. Schlick-Steiner BC, Steiner FM, Seifert B, Stauffer C, Christian Miralles A, Vences M (2013) New metrics for comparison of E, Crozier RH (2010) Integrative taxonomy: a multisource taxonomies reveal striking discrepancies among species approach to exploring biodiversity. Annual Review of Entomol- delimitation methods in Madascincus lizards. PLoS ONE, 8(7), ogy, 55, 421–438. e68242. Schoville SD, Roderick GK (2010) Evolutionary diversification Monacell JT, Cardone I (2014) Mobyle SNAP Workbench: a of cryophilic Grylloblatta species (Grylloblattodea: Grylloblat- web-based analysis portal for population genetics and evolu- tidae) in alpine habitats of California. BMC Evolutionary Biol- tionary genomics. Bioinformatics, 30, 1488–1490. ogy, 10, 163. Niemiller ML, Near TJ, Fitzpatrick BM (2012) Delimiting spe- Silvestro D, Michalak I (2012) raxmlGUI: a graphical front-end cies using multilocus data: diagnosing cryptic diversity in for RAxML. Organisms Diversity & Evolution, 12, 335–337. the southern cavefish, Typhlichthys subterraneus (Teleostei: Sipe TW, Browne RA (2004) Phylogeography of masked (Sorex Amblyopsidae). Evolution, 66, 846–866. cinereus) and smoky shrews (Sorex fumeus) in the southern O’Meara BC (2010) New heuristic methods for joint species delim- Appalachians. Journal of Mammalogy, 85, 874–885. itation and species tree inference. Systematic Biology, 59,59–73. Solıs-Lemus C, Knowles LL, Ane C (2015) Bayesian species O’Neill EM, Schwartz R, Bullock CT et al. (2012) Parallel delimitation combining multiple genes and traits in a unified tagged amplicon sequencing reveals major lineages and phy- framework. Evolution, 69, 492–507. logenetic structure in the North American tiger salamander Sousa V, Hey J (2013) Understanding the origin of species with (Ambystoma tigrinum) species complex. Molecular Ecology, 22, genome-scale data: modelling gene flow. Nature Review 111–129. Genetics, 14, 404–414. Opatova V, Arnedo MA (2014) Spiders on a hot volcanic roof: Stamatakis A (2006) RAxML-VI-HPC: maximum likelihood- colonisation pathways and phylogeography of the Canary based phylogenetic analyses with thousands of taxa and Island endemic trap-door spider Titanidiops canariensis (Ara- mixed models. Bioinformatics, 22, 2688–2690. neae, Idiopidae). PLoS ONE, 9, e115078. Stamatakis A (2014) RAxML version 8: a tool for phylogenetic Pons J, Barraclough TG, Gomez-Zurita J et al. (2006) Sequence- analysis and post-analysis of large phylogenies. Bioinformat- based species delimitation for DNA taxonomy of unde- ics, 30, 1312–1313. scribed insects. Systematic Biology, 55, 595–609. Stephens M, Donnelly P (2003) A comparison of Bayesian Potter KM, Frampton J, Josserand SA, Nelson CD (2008) methods for haplotype reconstruction from population Genetic variation and population structure in Fraser fir genotype data. American Journal of Human Genetics, 73, 1162– (Abies fraseri): a microsatellite assessment of young trees. 1169. Canadian Journal of Forest Research, 38, 2128–2137. Stephens M, Smith N, Donnelly P (2001) A new statistical Price EW, Carbone I (2005) SNAP. Workbench management method for haplotype reconstruction from population data. tool for evolutionary population genetic analysis. Bioinformat- American Journal of Human Genetics, 68, 978–989. ics Applications Note, 21, 402–404. Swofford DL (2002) PAUP*: Phylogenetic Analysis Using Parsi- Pritchard JK, Stephens M, Donnelly P (2000) Inference of popu- mony (*and Other Methods). Version 4. Sinauer Associates, lation structure using multilocus genotype data. Genetics, Sunderland, Massachusetts. 155, 945–959. Tamura K, Stecher G, Peterson D, Filipski A, Kumar S (2013) Pritchard JK, Wen X, Falush D (2010) Documentation for Struc- MEGA6: molecular evolutionary genetics analysis version ture software: Version 2.3. Available from http:// 6.0. Molecular Biology & Evolution, 30, 2725–2729. pritch.bsd.uchicago.edu/structure.html. 1-38. Thomas SM, Hedin M (2008) Multigenic phylogeographic Rambaut A, Suchard MA, Xie D, Drummond AJ (2014) Tracer divergence in the paleoendemic southern Appalachian opi- v1.6. Available from http://beast.bio.ed.ac.uk/Tracer lionids Fumontana deprehendor Shear (Opiliones, Laniatores,

Triaenonychidae). Molecular Phylogenetics and Evolution, 46, 645–658. Data accessibility Weisrock DW, Larson A (2006) Testing hypotheses of specia- Unphased DNA sequences: GenBank Accession nos tion in the Plethodon jordani species complex with allozymes KR607601–KR607977. and mitochondrial DNA sequences. Biological Journal of the Linnean Society, 89,25–51. Phased DNA sequence alignments, RAxML.tre files, Weisrock DW, Rasoloarison RM, Fiorentino I et al. (2010) morphology JPG files, POFAD matrices, STRUCTURE files, Delimiting species without nuclear monophyly in Madagas- transcriptome assemblies: Dryad doi:10.5061/ car’s mouse lemurs. PLoS ONE, 5, e9883. dryad.r87k3. Xie WG, Lewis PO, Fan Y, Kuo L, Chen MH (2011) Improving Illumina data submitted to NCBI Short Read Archive marginal likelihood estimation for Bayesian phylogenetic – Mt. Gibbes (SRX652506), Mt. Buckley (SRX1004559). model selection. Systematic Biology, 60, 150–160. Marker development information, voucher data, PCR Yang Z, Rannala B (2010) Bayesian species delimitation using multilocus sequence data. Proceedings National Academy of Sci- conditions and nuclear gene trees uploaded as online ences USA, 107, 9264–9269. Supporting Information. Yang Z, Rannala B (2014) Unguided species delimitation using DNA sequence data from multiple loci. Molecular Biology & Evolution, 31, 3125–3135. Zhang C, Zhang D-X, Zhu T, Yang Z (2011) Evaluation of a Supporting information Bayesian coalescent method of species delimitation. System- Additional supporting information may be found in the online ver- atic Biology, 60, 747–761. sion of this article. Zhang J, Kapli P, Pavlidis P, Stamatakis A (2013) A general species delimitation method with applications to phyloge- Material S1. Development of Nuclear Markers. netic placements. Bioinformatics, 29, 2869–2876. Material S2. Voucher number, location and GenBank information.

M.H. implemented the study design, directed data col- Material S3. Primers and gene annotations lection, conducted analyses and helped in writing the Material S4. Figure illustrating microgeographic mitochondrial manuscript. D.C. conducted analyses and contributed to structuring in the Black Mountains. writing the manuscript. F.C. collected specimens, assisted in study design and contributed to writing the Material S5. RAxML maximum likelihood nuclear gene trees. manuscript. Material S6. Marginal likelihood and Bayes Factor values for alternative species delimitation hypotheses, based on combined mitochondrial and nuclear data.